Town: client cells from their latest run; the backlog after a fresh village

decePub's e2e runs append to out/client.jsonl. The report kept the newest row per test and account, and the backlog
every row, so a failure from a run that picked other accounts stayed in both after later runs passed. Each client
cell now counts as its latest run left it (check.latest_client: the rows within an hour of its newest).

The backlog comes from a fresh village on today's build: 2556 checks pass and 4 fail, three of them peer-to-peer
counts, and one GoToSocial losing a status from its cache (an Update handled as a Create; a restart heals it),
recorded in INTEROP.md rather than as a gap, which would hide our own edits failing there. The community vote and the
Lemmy thread that failed before were data from older builds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw

tools/pasture/town/check.py

@@ -438,6 +438,21 @@ def canary_hits(platform, canaries):
 	return sorted(c for c in canaries if c in dump)
 
 
+def latest_client(lines):
+	"""decePub's e2e cells (out/client.jsonl, appended by every run) as each cell's latest run left them: the rows from
+	within an hour of the cell's newest, the newest per test. A run that picked other accounts leaves no stale failure."""
+	from datetime import datetime, timedelta
+	at = lambda c: datetime.fromisoformat(c["t"].replace("Z", "+00:00"))
+	newest = {}
+	for c in lines:
+		newest[c["cell"]] = max(newest.get(c["cell"], at(c)), at(c))
+	latest = {}
+	for c in sorted(lines, key=at):
+		if at(c) >= newest[c["cell"]] - timedelta(hours=1):
+			latest[(c["cell"], c.get("ref"))] = c
+	return list(latest.values())
+
+
 def load_gaps():
 	try:
 		with open(os.path.join(HERE, "gaps.json")) as f: